Research On Forecasting Of Steam Coal Price Based On Robust Regularized Kernel Regression Ensemble Method

Forcecasting of steam coal price is one of the key links in analyzing the steam coal market,and provides decision-making for energy enterprises in the steam coal market to make purchasing plans.The steam coal market is a complex and non-linear system,which contains macroeconomic factors,steam coal transportation,steam coal supply and demand.The influencing factors are characterized by a wide range,large amount of feature data and high noise.This topic uses the integrated model based on robust kernel regression to mine the price.law of steam coal from a large amount of steam coal price data,and the research contents are as follows:(1)A robust kernel regression model for forecasting the price of steam coal based on mixed kernel function is proposed to solve the problem that the nonlinear relationship in the steam coal dataset is difficult to model.The model uses a convex combination of Laplace kernels and polynomial kernels as mixed kernels.The Laplacian kernel function is a radial basis function that focuses on learning local rules of sample data.Polynomial kernel functions focus on learning the global rules of sample data.The mixed kernel function based on the convex combination of the two takes into account both local and global rules in the dataset.Selecting Huber loss function reduces the negative impact of anomalous noise in the dataset on model performance.Compared with single model,the proposed model is validated with steam coal dataset in terms of evaluation index RMSE,MAE and MAPE,and has the best evaluation index value.(2)To solve the problem that it is difficult for a single forecasting model to learn the rules of complex power coal data,an integrated forecasting model for steam coal price based on robust kernel regression is proposed.The empirical mode decomposition method is used to decompose the predictive error of robust kernel regression into N error components.Different differential autoregressive moving average models are used to fit and forecast the error components,respectively.The cumulative error component predictions are feed back to the robust kernel regression predictions to get the final price predictions.Error compensation is made by fiting robust kernel regression model prediction errors using a combination model based on empirical mode decomposition model and differential autoregressive moving average model to make up for the limitations of a single prediction model.The experimental results based on the steam coal data show that the integrated model has better prediction performance than the single model.(3)Research and development of steam coal price forecasting system.The steam coal price forecasting system can be divided into data source management and monitoring module,model building and training module,forecasting result display and model evaluation module.The data source management and monitoring module enables data selection and data reading.Model building and training module can implement data processing,model building,and model training functions.The prediction result display and model evaluation module can display the prediction result and evaluate the model performance.